Initial Solution Set Improvement for a GeneticAlgorithm in a Metadata Generation Support System for Landscape Photographs

摘要

In our metadata generation support system for landscape photographs, we use a genetic algorithm to find locations of photographs. Given a set of randomly generated solutions, the genetic algorithm tends to redundantly explore the search space because it is often that many worse solutions are distributed globally and a few better solutions are distributed locally in the search spaces of our search problems. To avoid such redundant searches, we propose a heuristic method to relocate worse solutions near better solutions before we execute the genetic algorithm. We show that the relocated initial solutions contribute to rinding better solutions than randomly generated solutions by an experiment.